The medallion architecture is a data design pattern that organizes lakehouse data into three progressively refined layers: Bronze holds raw ingested data, Silver holds cleaned and conformed data, and Gold holds business-ready aggregates. Data flows Bronze → Silver → Gold, getting more valuable and trustworthy at each step.
Last updated July 2026.
Bronze is the raw landing zone. You ingest source data — files, streams, database change feeds — with as little transformation as possible, keeping a faithful, append-only history of what came in.
Silver takes Bronze and makes it trustworthy: correct data types, deduplication, null handling, quality checks, and joins that resolve entities into a consistent, queryable model.
Gold is purpose-built for consumption: aggregated metrics, dimensional models, and feature tables shaped for specific BI dashboards, reports, or ML use cases.
Splitting the pipeline into layers gives you separation of concerns. Each stage has one job, which makes the whole system easier to debug, test, and trust:
On the Databricks lakehouse, each layer is typically a set of Delta Lake tables governed by Unity Catalog. Raw data lands in Bronze via COPY INTO or Auto Loader; declarative pipelines and Spark SQL/PySpark transform Bronze into Silver and Silver into Gold, with expectations enforcing data quality along the way. Because every layer is a Delta table, you get ACID transactions, time travel, and schema enforcement at each step.
Exam tip: Watch for questions that describe a table's role rather than naming it. "Raw as-landed" = Bronze, "validated and conformed" = Silver, "business-level aggregates" = Gold. Recognizing the role is often the whole question.
No — it's a recommended pattern, not a hard rule. It's widely adopted because it makes pipelines easier to reason about, but you can use fewer or more layers depending on your needs.
They're essentially the same idea. "Bronze" is the medallion term for the raw landing layer — data ingested as-is, usually with a little ingestion metadata added, before any cleaning.
Yes. Bronze/Silver/Gold is the standard three, but teams sometimes add sub-layers (for example, a raw vs. cleansed split within Silver). The principle — progressive refinement — matters more than the exact count.
Usually Gold, because it holds the business-ready aggregates shaped for reporting. Some ad-hoc analysis reads from Silver, which is clean and conformed but not yet aggregated.
The Bronze → Silver → Gold flow is one of the most-tested ideas on the exam. Work through it hands-on in the transformation chapters.
Open the Medallion chapter →